摘要：AbstractThe proposed modeling philosophy is calledimpulse response preserving (IRP)FIR modeling. It can be seen as a mixture of an OE and a FIR model combining the advantages and avoiding most drawbacks of both approaches. In the originally proposed approach via Gaussian processes a kernel needs to be established that expresses the prior knowledge about the process. The IRP approach is simpler, more straightforward, and much easier to understand, at least from a dynamic systems or controls point-of-view.The key prior information utilized by the IRP approach is the prior system ordern.Therefore the topic of order selection is also addressed. The IRP approach is significantly more robust w.r.t. the selected order. However, an extension of IRP that ensures plausible poles during optimization allows for a sensitive order selection.
